playwright-mcp-tabbed

中文说明

> A tab-aware Playwright MCP server for parallel agent workflows.

playwright-mcp-tabbed adds explicit tab_index support to Playwright MCP tools so multiple agents can operate on different tabs while sharing a single browser context and login session.

The official @playwright/mcp model is centered around a shared active page. That is perfectly fine for single-agent flows, but it becomes fragile in concurrent workflows:

• Agent A selects tab 1
• Agent B selects tab 2
• Agent A's next action may accidentally run on tab 2

This project removes that shared active-tab assumption. Instead, each tool call can target a tab directly.

playwright-mcp-tabbed is especially useful when browser automation is part of a larger agent workflow.

One main agent logs in once, opens several tabs, and assigns one tab_index to each bug-fixing sub-agent. Every sub-agent can reproduce and verify its own issue in parallel without losing authentication state.

After a refactor, different agents can validate /orders, /wallet, /settings, and /users at the same time while staying inside the same logged-in admin session.

One tab points to the old app, another to the migrated app, and another to a staging environment. Agents can compare behavior or styling in parallel without repeatedly logging in.

Instead of forcing one agent to serialize a long browser journey, you can split related subflows into dedicated tabs and assign them to separate agents.

• One browser instance
• One shared browser context
• Many tabs
• Every browser tool call can specify tab_index

This gives you:

• shared cookies and login state
• stable routing to the intended tab
• better fit for mul